CV

Professional Experiences

• July 2019 – March 2020: Distributed Coded Computation and Edge Learning Tsinghua University
  • Advisor: Professor Sheng Zhou
  • Considered network structure on the basis of distributed network.Proved that special network structure can effectively reduce the communication cost and reduce the influence of stragglers. Considered the minimum delay and minimum energy loss in a heterogeneous network with multiple base stations and multiple edge servers
  • Build a distributed computing architecture on Amazon EC2 and conduct relevant experiments.
• June 2020 – September 2020: Difference between Federated backdoor and centralized backdoor UChicago
  • Advisor: Neubauer Professor Ben Y. Zhao
    • Compare federal backdoor attacks with centralized backdoor attacks.
    • Determine whether current defenses against centralized backdoor attacks are effective against federal backdoor attacks.
• July 2020 – present : Coded Computation across heterogeneous Workers Tsinghua University
  • Advisor: PhD Candidate Yuxuan Sun
    • Consider introducing a queue system based on heterogeneous coded computation across heterogeneous workers
    • Consider how much more computing resources data coding consumes
• December 2020 – June 2021: Coded computation across shared heterogeneous workers with communication delay Tsinghua University
  • Advisor: Professor Sheng Zhou
    • Considering the randomness of communication and computation delays, we formulate a unified delay minimization problem for the joint allocation of computing power,communication bandwidth and task load
    • Use Markov’s inequality to simplify the problem, and transform the fractional worker assignment and resource allocation problem to max-min allocation by deriving its optimality condition. A greedy algorithm is proposed accordingly.
• August 2021 – November 2021: Stealing parameters from semi-supervised learning Helmholtz Center for Information Security
  • Advisor: Professor Yang Zhang
    • Consider the influence of different steal methods on different semi-supervise learning
    • Analyze the effects of data augmentation, loss function and data distributions on model stealing
    • Analyze the different importance of label and unlabel data in model stealing of semi-supervised learning
• September 2024 – Present: Protecting machine learning models from information leakage in edge-device deployments UCLA
  • Advisor: Professor Nader Sehatbakhsh
  • Addressed a critical security gap in on-device AI, demonstrating that TEE-shielded DNN partitioning can still leak information under large query budgets;
  • Developed an innovative method based on Fisher-information and LoRA that effectively blocks information leakage while maintaining full accuracy for authorized users.

Publications

• Fan Zhang, Yuxuan Sun, Sheng Zhou, Coded Computation over Heterogeneous Workers with Random Task Arrivals, in IEEE Communications Letters, 2021.
• Yuxuan Sun, Fan Zhang, Junlin Zhao, Sheng Zhou, Zhisheng Niu, Deniz Gunduz, Coded Computation across Shared Heterogeneous Workers with Communication Delay, in IEEE Transactions on Signal Processing, 2022.
• Fan Zhang, Peng Liu, Protecting the ‘Stop Using My Data’ Right through Blockchain-assisted Evidence Generation, Arxiv
• Fan Zhang, Ziqi Zhang, Hossein Khalili, Neiro Cabrera, Jonathan Xue, Nader Sehatbakhsh, FILOsofer: A TEE-Shielded DNN Partitioning Framework Based on Fisher Information-Guided LoRA Obfuscation, In submission.
• Hossein Khalili, Philip Do, Brandan Bright, Fan Zhang, Alexander Vilesov, Kittipat Apicharttrisorn, Nader Sehatbakhsh, Cloak of Invisibility: Real-Time Privacy-Preserving Volumetric Video Streaming, In submission.

Competitions and Awards

• Augest 2018 – December 2018: Team leader of The 20th Electronic Design Competition
  • Code on STM32 microcontroller to control the trolley.Use infrared sensor and ultrasonic device to aviod structure and use wifi system to abtain the location information from the host computer
  • Won The Winning Award in 67 teams
• March 2019 – May 2019: Member of The Second Artificial Intelligence Challenge Competition
  • Submit four AI codes for each team to control four characters, cooperate with each other to kill all non-teammates in the map
  • Entered The Finals in 127 teams
• September 2019 – December 2019: Member of The Practice of Intelligent Robot Design
  • Write code to control the robot to avoid obstacles, cross the gully, identify traps and identify traffic lights.Simultaneously control the robot’s steering gear to adjust the motion
  • Won the 3’rd Place Award and the Best Performance Award